ncnn

1
// Tencent is pleased to support the open source community by making ncnn available.
2
//
3
// Copyright (C) 2019 THL A29 Limited, a Tencent company. All rights reserved.
4
//
5
// Licensed under the BSD 3-Clause License (the "License"); you may not use this file except
6
// in compliance with the License. You may obtain a copy of the License at
7
//
8
// https://opensource.org/licenses/BSD-3-Clause
9
//
10
// Unless required by applicable law or agreed to in writing, software distributed
11
// under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR
12
// CONDITIONS OF ANY KIND, either express or implied. See the License for the
13
// specific language governing permissions and limitations under the License.
14

15
#version 450
16

17
#if NCNN_fp16_storage
18
#extension GL_EXT_shader_16bit_storage: require
19
struct sfpvec8 { f16vec4 abcd; f16vec4 efgh; };
20
#endif
21
#if NCNN_fp16_arithmetic
22
#extension GL_EXT_shader_explicit_arithmetic_types_float16: require
23
#endif
24

25
layout (constant_id = 0) const int storage_type_from = 0;
26
layout (constant_id = 1) const int storage_type_to = 0;
27

28
#define shape_constant_id_offset 2
29
layout (constant_id = shape_constant_id_offset + 0) const int dims = 0;
30
layout (constant_id = shape_constant_id_offset + 1) const int w = 0;
31
layout (constant_id = shape_constant_id_offset + 2) const int h = 0;
32
layout (constant_id = shape_constant_id_offset + 3) const int c = 0;
33
layout (constant_id = shape_constant_id_offset + 4) const int cstep = 0;
34

35
layout (constant_id = shape_constant_id_offset + 5) const int outdims = 0;
36
layout (constant_id = shape_constant_id_offset + 6) const int outw = 0;
37
layout (constant_id = shape_constant_id_offset + 7) const int outh = 0;
38
layout (constant_id = shape_constant_id_offset + 8) const int outc = 0;
39
layout (constant_id = shape_constant_id_offset + 9) const int outcstep = 0;
40

41
layout (binding = 0) readonly buffer bottom_blob { sfpvec8 bottom_blob_data[]; };
42
layout (binding = 1) writeonly buffer top_blob_fp32 { mat2x4 top_blob_fp32_data[]; };
43
#if NCNN_image_shader
44
layout (binding = 2) uniform unfp sampler3D bottom_blob_3d;
45
layout (binding = 3, rgba32f) writeonly uniform highp image3D top_blob_3d_fp32;
46
#endif
47

48
layout (push_constant) uniform parameter
49
{
50
    int dims;
51
    int w;
52
    int h;
53
    int c;
54
    int cstep;
55

56
    int outdims;
57
    int outw;
58
    int outh;
59
    int outc;
60
    int outcstep;
61
} p;
62

63
void main()
64
{
65
    int gx = int(gl_GlobalInvocationID.x);
66
    int gy = int(gl_GlobalInvocationID.y);
67
    int gz = int(gl_GlobalInvocationID.z);
68

69
    if (gx >= psc(outw) || gy >= psc(outh) || gz >= psc(outc))
70
        return;
71

72
    afpvec8 v;
73

74
    if (psc(dims) == 1)
75
    {
76
        if (storage_type_from == 0)
77
        {
78
            int gi = gx;
79

80
            v = buffer_ld8(bottom_blob_data, gi);
81
        }
82
#if NCNN_image_shader
83
        if (storage_type_from == 1)
84
        {
85
            v = image3d_ld8(bottom_blob_3d, ivec3(gx, 0, 0));
86
        }
87
#endif
88

89
        if (storage_type_to == 0)
90
        {
91
            int gi = gx;
92

93
            top_blob_fp32_data[gi] = mat2x4(v);
94
        }
95
#if NCNN_image_shader
96
        if (storage_type_to == 1)
97
        {
98
            int x2 = gx * 2;
99

100
            imageStore(top_blob_3d_fp32, ivec3(x2 + 0, 0, 0), v[0]);
101
            imageStore(top_blob_3d_fp32, ivec3(x2 + 1, 0, 0), v[1]);
102
        }
103
#endif
104
    }
105
    else if (psc(dims) == 2)
106
    {
107
        if (storage_type_from == 0)
108
        {
109
            int gi = gy * psc(w) + gx;
110

111
            v = buffer_ld8(bottom_blob_data, gi);
112
        }
113
#if NCNN_image_shader
114
        if (storage_type_from == 1)
115
        {
116
            v = image3d_ld8(bottom_blob_3d, ivec3(gx, gy, 0));
117
        }
118
#endif
119

120
        if (storage_type_to == 0)
121
        {
122
            int gi = gy * psc(outw) + gx;
123

124
            top_blob_fp32_data[gi] = mat2x4(v);
125
        }
126
#if NCNN_image_shader
127
        if (storage_type_to == 1)
128
        {
129
            int x2 = gx * 2;
130

131
            imageStore(top_blob_3d_fp32, ivec3(x2 + 0, gy, 0), v[0]);
132
            imageStore(top_blob_3d_fp32, ivec3(x2 + 1, gy, 0), v[1]);
133
        }
134
#endif
135
    }
136
    else // if (psc(dims) == 3)
137
    {
138
        if (storage_type_from == 0)
139
        {
140
            int gi = gz * psc(cstep) + gy * psc(w) + gx;
141

142
            v = buffer_ld8(bottom_blob_data, gi);
143
        }
144
#if NCNN_image_shader
145
        if (storage_type_from == 1)
146
        {
147
            v = image3d_ld8(bottom_blob_3d, ivec3(gx, gy, gz));
148
        }
149
#endif
150

151
        if (storage_type_to == 0)
152
        {
153
            int gi = gz * psc(outcstep) + gy * psc(outw) + gx;
154

155
            top_blob_fp32_data[gi] = mat2x4(v);
156
        }
157
#if NCNN_image_shader
158
        if (storage_type_to == 1)
159
        {
160
            int x2 = gx * 2;
161

162
            imageStore(top_blob_3d_fp32, ivec3(x2 + 0, gy, gz), v[0]);
163
            imageStore(top_blob_3d_fp32, ivec3(x2 + 1, gy, gz), v[1]);
164
        }
165
#endif
166
    }
167
}
168